Click on “Download PDF” for the PDF version or on the title for the HTML version.

If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.

Abstract. Sustainable weed management necessitates herbicide-reduced precision weed control. Recent advancements in machine vision and artificial intelligence for weed recognition and localization open up opportunities for enhanced precision weeding. Although blanket spraying systems are widely implemented for weed removal, the development of vision-guided, smart sprayers has become increasingly important but remains to be fully explored for many vegetable cropping systems. This study presents a new engineering effort to develop vision-guided, smart precision weeding technology for vegetable crops. A new ground-based, fully integrated sprayer prototype was designed and assembled, consisting of a machine vision system for weed detection and a weeding actuator built with an array of 12 evenly spaced spraying nozzles controlled independently, covering the row width of about 0.66 m. A graphical user interface was designed and implemented for system integration and operation. Built on a diverse, three-season weed dataset for the detection of Lambsquarters and Purslane, YOLOv10s yielded an overall mAP@50 of 87.5% in the offline test with still images. With the model deployed onboard, the sprayer prototype, operated at a platform speed of about 0.11m/s, was subjected to experimental trials in both indoor and field conditions for real-time target detection and detection-based spraying. In detecting simulated targets in the indoor testing, the prototype attained a video detection accuracy of 97.67% and a spraying hit rate of 88.35%. For the testing in real field conditions, the performance deteriorated with a video-based weed detection accuracy of 85.52% and a spraying hit rate of 78.43%. The imperfection in nozzle configurations and the relatively large nozzle spacing (about 51 cm) contributed primarily to off-target spraying, and the timing errors for nozzle actuation also affected the spraying efficacy. Further improvements in both hardware and software as well as extensive field tests are needed for developing the protype into a reliable, high-resolution sprayer for vegetable weeding.